Cursor Marketplace Attack Traced to Fireworks AI Access Key
TL;DR - Cursor says the incident started after an access key issued to Fireworks was exposed. - The attacker used exposed credentials to access selected Cursor systems and publish malicious Claude Code extension copies. - Anthropic.claude-code versions 2.1.312 and 2.1.313 were available through the Cursor marketplace for approximately 11 hours on October 29. - At most 5,000 users who requested specific cached versions potentially received the malicious build. - Cursor says affected users should use its indicators and remediation steps because uninstalling the extension alone may not be enough.
What happened#
Cursor’s investigation says the attack started at Fireworks. An attacker obtained an access key Cursor had issued to Fireworks. Cursor says the key was exposed by Fireworks, allowing the attacker to copy data from a read-only storage bucket used to share data with the company.
The bucket contained limited coding agent traces, mostly from 2025, from sessions not in Privacy Mode. Cursor says a small number of those traces contained credentials that had not been redacted, and that less than 0.01% of customers had API credentials exposed.
Using those credentials, the attacker reached selected SaaS tools used by Cursor. Cursor says the accessed information included selected archived non-production source code repositories, limited customer support data and email conversations, product usage metrics and contact details, summaries of some coding agent conversations, and some credentials users had sent to the Cursor support team. Cursor’s incident update says its monitoring identified the unauthorized activity, after which it deactivated the vendor key and began revoking and rotating credentials.
On October 29, the attacker used a stolen credential to place malicious, unauthorized copies of Anthropic’s Claude Code extension on the cache serving the Cursor VS Code Extension Marketplace. Cursor says the malicious copies were not published by Anthropic, and that Open VSX never contained the malware.
The affected packages were Anthropic.claude-code versions 2.1.312 and 2.1.313 for Windows and macOS. Cursor’s security notice says they were available between about 2:27am PT and 1:35pm PT on October 29, 2026. Cursor later said the malicious build was available for approximately 11 hours and that at most 5,000 users who requested specific cached versions potentially received it.
The malicious extension attempted to collect host and user information, add an attacker-controlled SSH key, enable the local SSH service, open a reverse SSH tunnel, and establish persistence. Cursor says the attacker server was taken offline with help from its cloud provider.
Are you affected#
Potentially affected users are those who installed or updated to Anthropic.claude-code 2.1.312 or 2.1.313 from the Cursor VS Code Extension Marketplace on macOS or Windows between 2:27am PT and 1:35pm PT on October 29, 2026.
Cursor says users in that group should check for the svc-telemetry SSH key and other indicators listed in its security notice. The notice includes modified VSIX hashes, the attacker endpoint, SSH fingerprints, persistence locations, process names, and network indicators.
Cursor specifically warns that uninstalling the extension alone may not be sufficient. Its notice says the payload drops standalone persistence scripts that may continue running. It recommends isolating hosts showing evidence of payload execution or backdoor installation, removing the unauthorized key and persistence mechanisms, assessing whether rebuilding is appropriate, and rotating credentials and sessions accessible from affected hosts.
Cursor says users who did not receive a notification about exposed coding agent traces were not included in those traces. It also says Privacy Mode prompts and code, active inference requests, customer code repositories, production systems and infrastructure, official Cursor updates, and other extensions were not affected.
What is confirmed and what is reported#
| Claim | Source | Status |
|---|---|---|
| The incident started after an access key issued to Fireworks was exposed. | Cursor | Confirmed |
| The attacker copied data from a read-only storage bucket used to share data with Fireworks. | Cursor | Confirmed |
| Less than 0.01% of customers had API credentials exposed. | Cursor | Confirmed |
| Malicious Anthropic.claude-code versions 2.1.312 and 2.1.313 were available through the Cursor marketplace. | Cursor | Confirmed |
| The malicious versions were available between about 2:27am PT and 1:35pm PT on October 29, 2026. | Cursor | Confirmed |
| At most 5,000 users potentially received the malicious build. | Cursor | Confirmed |
| Anthropic was not the publisher of the malicious copies, and Open VSX never had the malware. | Cursor | Confirmed |
| The attack began after Fireworks was compromised twice, involving internal credentials, stored secrets, account and billing information, a cloud provider, development machines, and customer data. | SecurityWeek | Reported |
| Notion and Uber also disclosed data breaches stemming from the Fireworks compromise. | SecurityWeek | Reported |
No CVE has been published in the notices cited here.
What defenders should do now#
- Review extension installation and update records for Anthropic.claude-code versions 2.1.312 and 2.1.313 on Windows and macOS during the exposure window.
- Use Cursor’s security notice to search endpoint, network, and extension telemetry for the listed hashes, endpoint, SSH key, fingerprints, files, persistence mechanisms, and processes.
- If the
svc-telemetrykey or other payload evidence is found, isolate the host and investigate for the unauthorized SSH key, persistence mechanisms, and running tunnel processes. - Follow Cursor’s platform-specific remediation steps. Cursor provides scripts for macOS and Windows and says they remove the malicious extension, persistence mechanisms, running tunnel processes, and attacker SSH key.
- Rotate credentials and sessions accessible from an affected host, using a trusted system. Cursor says limited environment-variable collection does not rule out later credential access through the backdoor.
- Assess whether an affected host should be rebuilt, particularly where there is evidence of payload execution or backdoor installation.
- Review notifications from Cursor. The company says it contacted affected customers directly and emailed team administrators lists of specific keys to rotate when credentials were exposed.
- Continue monitoring for anomalous activity. Cursor says it believes the incident is contained, and that the last attacker activity it observed was on November 1 and was blocked.
This advisory is informational and is updated as vendors publish details. Confirm versions and actions against the vendor’s own guidance for your environment.
Sources
- Primary notice: cursor.com
- Primary notice: cursor.com
- Research: securitysummits.com
- Research: virtual.cisoforum.com
- Report: securityweek.com